• Advertisement
Sign in to follow this  

Should NPCs learn in every game?

This topic is 4256 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Sorry I don't have a deep and lengthy question. [smile] I've been wondering should all games have the type of AI (or something similar) that Virtua Fighter 4. Or should that stay with games like Halo, VF4, and Ninja Gaiden?

Share this post


Link to post
Share on other sites
Advertisement
No. The fact you used every in the title such dictates my answer.

Why? Because sometimes I want an easy game. Or if you want to introduce a player to a game, you don't want the AI to scale up drastically too quickly.

And some games you are more about reflexes and the skill required to pull off moves, even when you know what's coming. Ever tried playing Street Fighter 2 on super hard mode where the enemies could cancel special tactics, or throw multiple fireballs at once? Now, imagine trying to play if they could learn how to counteract whatever you do. If the game techincally has a conclusion, a good game would be one where it's winnable. I do think AI has it's place, but the ability to learn should be limited to few genres.

Share this post


Link to post
Share on other sites
"Learning AI", which is what it'll be called now, doesn't have learn every move you do and whuppass in an instance. You CAN adjust the rate of learning or even if it learns at all. You make it sound like everything has to be super hard from the get go. Which is not what I'm suggesting [smile]

Share this post


Link to post
Share on other sites
Quote:
Original post by Nytegard
. . . I do think AI has it's place, but the ability to learn should be limited to few genres.


Agreed. While it can be very interesting to have an NPC seemingly "grow" while you play the game, it should only happen if it adds to the overall game experience. What if the player only interacts with that NPC a few times in the whole game, then it would be a serious waste of coding time for something that would have a minimal affect on the overall game. So it really depends on the type of game and how much of a role you want your NPC's to play. A vendor in an RPG type game would benefit from this because it would add depth to the game instead of just a place to dump loot.

Share this post


Link to post
Share on other sites
Virtua fighter has the best fighting AI I have seen.
They started stuiped, but slowly they are becoming more and more challenging, so it doesnt matter how good you are you will get all the levels of challege from very easy to the very hard and nigh impossible. And it is also very satisfying.
I have played days in virtual fighter 1 and 2. These are the best 1 on 1 fighting games ever. Too bad not many people are aware to that.
Of course if it gets too hard, you can always delete all the AI of the NPC in virtua fighter, and start all over again.

Share this post


Link to post
Share on other sites
Guest Anonymous Poster
Lost real people dont learn very fast (or seemingly at all).

Depending on the game, you might have most of the NPCs keep their original abilities (and the waves of ordinary opponents can have their abilities set appropriately for the specific situations (easy areas, hard areas, areas with a specific theme, etc...)

You might then have only a few that develop in parallel with the player (ie- a nemesis/arch enemy who is repeated confronted at different points, or the players companion NPCs who may evolve/improve).

The NPCs advancement doesnt nessessarily have to be linked directly to involvement with thye player either.

Share this post


Link to post
Share on other sites
One other problem you're going to run into in many games is that the average enemy doesn't live long enough to learn from the player's fighting style (and wouldn't have any in-game reason to know how the player fought previous enemies). So if you make a learning AI, it'll be giving enemies knowledge that they really shouldn't have, which ends up meaning that it cheats.

Also, some games really aren't so much about the player competing against the AI as they are about the player competing against themselves. Racing games, for example, after a point tend to be more about the player trying to beat their previous best time than they are about the player beating the other racers (though players don't just do time trial because e.g. they want the "moving obstacles" on the track). Basically any scored game will ultimately stop being "Can I win at all?" and turn into "Can I win better than I won last time?" Those kinds of games should not have adaptive AI, because that would remove the ability of the player to fairly compare any two runs of the game.

Share this post


Link to post
Share on other sites
Quote:
Original post by Alpha_ProgDes
"Learning AI", which is what it'll be called now, doesn't have learn every move you do and whuppass in an instance. You CAN adjust the rate of learning or even if it learns at all. You make it sound like everything has to be super hard from the get go. Which is not what I'm suggesting [smile]
Also, learning AI doesn't have to mean "learns to win". On the contrary, "very easy" might have AI that "learns to lose" such that it makes the types of mistakes that the player has learned to take advantage of.
I still wonder why more games don't have this kind of thing. I suppose, though, games are just barely starting to having decent AI, so expecting them to grow in both directions might be a bit much.

Share this post


Link to post
Share on other sites
Quote:
Original post by Extrarius
Quote:
Original post by Alpha_ProgDes
"Learning AI", which is what it'll be called now, doesn't have learn every move you do and whuppass in an instance. You CAN adjust the rate of learning or even if it learns at all. You make it sound like everything has to be super hard from the get go. Which is not what I'm suggesting [smile]
Also, learning AI doesn't have to mean "learns to win". On the contrary, "very easy" might have AI that "learns to lose" such that it makes the types of mistakes that the player has learned to take advantage of.
I still wonder why more games don't have this kind of thing. I suppose, though, games are just barely starting to having decent AI, so expecting them to grow in both directions might be a bit much.

Probably we need a parallelized AI card capable 1.8GFlops.... [shakeshead]

In all seriousness, with all the cards that are dual processor and dual/quad core, I'm surprised there hasn't been a sudden surge in AI, as opposed to physics. I guess we're not ready to "twitch think"* yet.


*my word. i called it. "twitch think" is trademarked (2006) by ME!

Share this post


Link to post
Share on other sites
Quote:
Original post by Extrarius
Quote:
Original post by Alpha_ProgDes
"Learning AI", which is what it'll be called now, doesn't have learn every move you do and whuppass in an instance. You CAN adjust the rate of learning or even if it learns at all. You make it sound like everything has to be super hard from the get go. Which is not what I'm suggesting [smile]
Also, learning AI doesn't have to mean "learns to win". On the contrary, "very easy" might have AI that "learns to lose" such that it makes the types of mistakes that the player has learned to take advantage of.
I still wonder why more games don't have this kind of thing. I suppose, though, games are just barely starting to having decent AI, so expecting them to grow in both directions might be a bit much.


Take a look at UT2003/4. Their AI which learned to win or lose could be horribly exploited. While, I agree that maybe something in a single player campaign shouldn't matter, when you start introducing multiplayer capabilities into the mix, this can cause a problem. Not all people have the same abilities or skill, and if player A is awesome, player B's turn might not work out so well. On the other hand, being as how one of the key factors of a multiplayer game is to one up other players so to speak, if player A wins the race in 1 min, making player B & C have a tougher time, should player B be working with C, he or she could purposely throw the game and look real bad, so player C has an incredibly easy time. I guess my main concern as I originally stated is that every game is too broad to make this work.

Share this post


Link to post
Share on other sites
Guest Anonymous Poster
If "Learning AI" is to be used, I believe the way an NPC should behave all depends on the type of game/setting - it depends on the artificial universe that is described/portrayed and the imaginary rules in which the NPCs would (should) be constricted to. I think the question the developer must ask himself is: "If I was this character, with these abilitys, living in this imaginary game world; How would I think/react??"

The answer to that question is what you are attempting to simulate... just dont make it too hard then no one would play =P

Share this post


Link to post
Share on other sites
Guest Anonymous Poster
Quote:
Original post by Alpha_ProgDes
Quote:
Original post by Extrarius
Quote:
Original post by Alpha_ProgDes
"Learning AI", which is what it'll be called now, doesn't have learn every move you do and whuppass in an instance. You CAN adjust the rate of learning or even if it learns at all. You make it sound like everything has to be super hard from the get go. Which is not what I'm suggesting [smile]
Also, learning AI doesn't have to mean "learns to win". On the contrary, "very easy" might have AI that "learns to lose" such that it makes the types of mistakes that the player has learned to take advantage of.
I still wonder why more games don't have this kind of thing. I suppose, though, games are just barely starting to having decent AI, so expecting them to grow in both directions might be a bit much.

Probably we need a parallelized AI card capable 1.8GFlops.... [shakeshead]

In all seriousness, with all the cards that are dual processor and dual/quad core, I'm surprised there hasn't been a sudden surge in AI, as opposed to physics. I guess we're not ready to "twitch think"* yet.


*my word. i called it. "twitch think" is trademarked (2006) by ME!





Intel has plans for CPUs with as many as 16 full cores. I hope that the ability to keep them all fed with data keeps up (AI is often heavily random data access and will require bigger data sets than the 2 MB cache they plan to give each core).

We will need that kind of increase too. Real AI can easily gobble up magnitudes more processing power (and I would hope that the Game Worlds would also be getting more complex -- more objects with behaviors/interactions -- increasing twhat the AI has to contend with....)

The games have only begun to make use of multiple cores as only recently have they begun to go mainstream.


Hah Hah, my 'Fast Think'(tm 2005) and 'Smartz'(tm 2003) terms will stomp on your wee term.......




Share this post


Link to post
Share on other sites
Quote:
Original post by Nytegard
I guess my main concern as I originally stated is that every game is too broad to make this work.


[sarcasam]What!? Dont'cha know that learning AI is just what we need to improve TETRIS! [/sarcasam]

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement